Memory

Durable memory with local keyword and semantic recall.

BudgetAI remembers useful context between sessions using a main MEMORY.md file plus dated notes in Markdown. It recalls preferences, project notes, and task traces without forcing every old transcript into the active prompt.

Search modes

  • Keyword — SQLite full-text search.
  • Semantic — sqlite-vec powered by on-device ONNX embeddings, so memory never leaves your machine. You can switch embeddings to OpenAI or Ollama.
budgetai memory search "deployment steps"
budgetai memory add "Prefers pnpm over npm"

Optional behaviors

  • Retention — older memories can slowly fade.
  • Dream mode — an opt-in pass that cleans up and merges memories.
budgetai configure memory-embedding
budgetai memory flush --status

Semantic recall runs locally by default. No memory content is sent to a remote model unless you explicitly choose a remote embedding provider.

PreviousTool CompressionNext Skills